3DAi Marketing Solutions
Powered by Trezdev
← Back to Dashboard
Overview Session Protocol Folder Structure Agent Team Tool Stack 🧠 The Brain Active Clients Rules ⚡ The Machine
Built by Kevin Castro × Claude
Real marketing skill, paired with AI — and an entire marketing company built to run itself.
This entire system was built by Kevin Castro, CMO of Trezdev — happy to bring his marketing skill together with Claude, Anthropic's frontier AI, to create it. Not a template, not off-the-shelf software: a living, multi-agent operation architected piece by piece — the agents, the rules, the pricing engine, the client pipeline, and the dashboards your partners are looking at right now.
The result is a real company that runs like a team of specialists — content, creative, scheduling, analytics, websites, pricing, and strategy — each one an AI agent that knows its job and hands off to the next. This is what one person, working alongside AI, can now build.
⚡ 8 AI agents, one coordinated team
🏗️ Built from zero — fully custom
🚀 Live and running real clients
What you're looking at: Trezdev runs a team of 8 AI agents — each one has a specific job, like a real employee. They create content, publish to social media, pull analytics, build websites, and run strategy. This document explains exactly how that system is organized, what files control it, and how it stays coordinated across multiple clients — without any agent getting confused or mixing things up.

The simple version: Think of this like a company with 8 employees. Each employee has a job description (their AGENT.md file), a company handbook they all follow (CLAUDE.md), a shared office whiteboard (STATUS.md), and a client folder for each account they work on (CLIENT-TRUTH.md). Every time someone starts a new work session, they read the handbook, check the whiteboard, and open the right client folder before doing anything.
How It Works — The Big Picture
Trezdev's system is built around one core idea: every agent knows exactly what to read, in what order, before touching anything. This prevents cross-client confusion, keeps all agents in sync, and makes the whole system portable — it runs on Kevin's local machine today and can move to a server tomorrow without changing a single file.

There are three layers of files every agent works with:

Layer 1 — Universal (same for everyone, every session): CLAUDE.md + AGENCY-TRUTH.md. These never change per client. They define the company, the rules, and the tools.

Layer 2 — Agency-wide state: ops/STATUS.md. The shared whiteboard. Tells every agent which clients are active and what's happening across the whole agency right now.

Layer 3 — Client-specific: clients/{client}/STATUS.md + clients/{client}/CLIENT-TRUTH.md. The only place client facts and session notes live. Each client has their own isolated folder. Agents are only allowed to read and write to the ONE client they're working on in that session.
Every time an agent is loaded — whether it's Mary writing a content brief or Parker scheduling posts — they run through these exact steps before doing any work. There are no exceptions. This is what prevents an agent from accidentally writing VoiceLink content into a Home Inspection folder, or publishing to the wrong client's social accounts.

Step 0 is the most important: if Kevin doesn't specify which client, the agent stops completely and asks. No file gets read, no work gets done until the client is confirmed. This is the first line of defense against cross-client errors.
These steps run in order before any work begins. No exceptions.
0
IF no client name provided → STOP
Ask Kevin: "Which client are we working on today?" Do not read any files or do any work until Kevin confirms the client name. This is non-negotiable — it prevents any possibility of mixing up client files.
1
agents/{my-name}/AGENT.md
The agent reads their own job description. This file defines exactly what they own, what they don't touch, and their locked behaviors. It never changes per client — it's the same rules whether working on VoiceLink or Home Inspection.
2
AGENCY-TRUTH.md
The company tool bible. Every MCP tool, endpoint, publishing rule, and infrastructure fact lives here. If a tool isn't listed here, the agent doesn't use it. This ensures every agent is using the same tools the same way, every time.
3
ops/STATUS.md
The agency whiteboard. The agent checks which clients are active, any system-wide alerts or flags, and what other agents last reported. This is the agency-level "what's going on right now" check.
4
Marketing Agency/clients/{client}/STATUS.md
The client-specific whiteboard. What every agent last did for THIS client. What's in progress. What's blocked. What's next. This is how agents hand off work to each other — without ever having a live conversation.
5
Marketing Agency/clients/{client}/CLIENT-TRUTH.md
The single source of truth for this specific client. Platforms they publish on, account IDs, posting times, brand voice, content pillars, banned words, overlay placement rules. Every client has their own version of this file. This is what makes an agent "know" the client — not memory, but facts locked in a file.
6
CONFIRM — Required before any work begins
The agent says out loud exactly what they loaded. This is an accountability checkpoint — it creates a paper trail and forces the agent to verify they read the right files for the right client.
"I am [agent name] working on [client name]. I have read [client]/STATUS.md and [client]/CLIENT-TRUTH.md. Ready to proceed."
7
Work
Only now does the agent actually do their job — brief writing, image generation, publishing, analytics, etc.
8
Write to Marketing Agency/clients/{client}/STATUS.md
Before closing, the agent logs what they did, what's next, and anything that's blocked. This is how the next agent — or this same agent in a future session — picks up exactly where things left off. No verbal handoffs. No guessing. It's all in the file.
9
CLOSING CONFIRM — Wait for Kevin
The agent confirms exactly which file they saved to and waits for Kevin to acknowledge before ending the session. This prevents silent saves to the wrong file.
"Session complete. I wrote to clients/[client name]/STATUS.md. No other client files were touched. Please confirm."
Do not close until Kevin confirms.
There are 6 critical file types in this system. Knowing what each one is for — and more importantly, what it is NOT for — is what keeps the whole system clean. Every file has exactly one job. No file doubles up.
Shared — All Agents, All Clients
CLAUDE.md
The company handbook. Every agent reads this at the start of every single session — before doing anything else. It defines what Trezdev is, what products exist, the 10-step session protocol, the folder structure, and the non-negotiable rules. Think of it like the employee handbook that every new hire reads on day one — except every employee re-reads it every single morning.
Read by: Every agent, every session
Shared — All Agents, All Clients
AGENCY-TRUTH.md
The tool room. Contains every software tool the agency uses, how to use it, what it's for, and what's been retired. MCP endpoints, model names, publishing rules, deprecated tools — all here, nowhere else. If an agent wants to publish a post or generate an image, they come here first to confirm exactly which tool to use and how. No agent ever guesses at tools.
Read by: Every agent, every session
Ops — Agency-Wide Only
ops/STATUS.md
The agency whiteboard. When an agent starts a session, they check this to see which clients are active, any system-wide warnings (like a tool being broken or a platform having a character limit), and what Kasey last reported. This file stays clean — it only holds agency-level information, never client-specific session notes. Client notes go in the client's own STATUS.md.
Read by: Every agent, every session
Per Client — Isolated
clients/{client}/STATUS.md
The client's daily whiteboard. Every agent writes here at the end of their session — what they did, what they built, what's ready, what's blocked, what needs to happen next. When the next agent loads for this client, they read this file and know exactly where things stand. This is how 8 separate agents stay in sync on the same client without ever talking to each other directly.
Written by: All agents after every session
Per Client — Single Source of Truth
clients/{client}/CLIENT-TRUTH.md
Everything true about this specific client — and the ONLY place it lives. Social media platforms they use, account IDs for each platform, posting schedule, brand voice, content pillars, banned words, logo placement, CTA rules. Every agent reads this before doing any client work. If a fact about a client changes, it gets updated here and here only — never in multiple places.
The single source of truth — no duplicates anywhere
Per Agent — Their Job Description
agents/{name}/AGENT.md
Each agent's personal instruction file. Defines their exact role, what they own, what they are forbidden from touching, and their locked behavioral rules. Mary's AGENT.md tells her how to write briefs. Cameron's tells him exactly how to go through his 4 generation gates. These files never change per client — the agent is the same person regardless of which client they're working on.
Read by: That agent only, every session
The system is designed so that adding a new client requires zero changes to any shared file or any agent's instructions. The agents already know how to work with any client — they just need the client's folder and facts. Here's the exact process:
1
Copy _template/ → clients/{new-client-slug}/
The template folder has the exact right structure — all subfolders, all placeholder files. Every new client starts from the same clean base.
2
Fill in CLIENT-TRUTH.md completely
Company name, contact info, social platforms, brand voice, content pillars, posting schedule, banned words, CTA rules. Leave nothing blank — agents rely entirely on this file to know the client.
3
Connect social accounts in Zernio
Add each platform account in Zernio's dashboard. Copy the account IDs back into CLIENT-TRUTH.md. Parker needs these IDs to schedule posts to the right accounts.
4
Add client to ops/STATUS.md active client index
Every agent checks this table to know which clients exist and where their files are. A client not listed here is invisible to the system.
5
Brief Mary — she does the rest
Load Mary with the new client name. She reads CLIENT-TRUTH.md, writes the first 8-week content strategy, and briefs Cameron. From there the full agent cycle runs exactly as it does for every other client.
All 8 agents are now ready to work on this client. No additional setup needed.
From the first prospect meeting to a live campaign generating leads — here is every agent, every handoff, and every step in the system. Follow the arrows.
9
Specialized Agents
4
Onboarding Weeks
8
Campaign Weeks
13+
Weeks to Ad Spend
Every agent has one job. Every handoff is documented. Nothing moves until the previous step is confirmed complete. This is how the machine never breaks.
The Team — 9 Agents
The Team · 9 Agents
Meet the Agents
Every agent has one job. No overlaps. No guessing. Each owns their domain completely.
D
Dakota
Client Onboarding & Strategy
Runs every new client through discovery. Fills CLIENT-TRUTH.md. Creates folder structure. Briefs all agents. The first agent a client ever touches.
J
Jim
Pricing & Proposals
Takes Dakota's discovery answers and Mary's content plan. Outputs a three-part proposal: setup fee, monthly retainer, and ad spend recommendation. Same day.
J
Judy
Website & Dashboards
Builds every client dashboard, Onboarding page, and the agency hub. The architect. Owns all HTML structure — never content.
M
Mary
Content Strategy
Writes the weekly brief. Assigns hooks, tracks, platforms, and fire times. Uses Eli's analytics to sharpen every next week. The brain of the content machine.
C
Cameron
Content Production
Turns Mary's brief into finished content. AI images, AI video, compositing, captions. Every post goes through 4 gates before Cameron hands it to Parker.
P
Parker
Publishing & Scheduling
Takes Cameron's finished posts. Runs pre-flight checks. Schedules everything in Zernio at Mary's exact fire times. Verifies every post went live. Never asks Kevin to check.
E
Eli
Analytics & Learning
Pulls performance data Wednesday and Saturday. Updates the analytics dashboard. Delivers a full report to Mary every Saturday EOD. The data that makes the content smarter.
R
Riley
Repurposing Engine
Takes one piece of approved content and turns it into 30+ derivatives. Every winner gets stretched across every format and platform. Nothing goes to waste.
Phase 1 · Weeks 1–4 · Onboarding
Phase 1 · Onboarding · Weeks 1–4
New Client Comes In — Here's What Happens
Follow the path. Every step triggers the next. Nothing is skipped.
🌱 Stage 0 · Research / Prospect Before any formal discovery — get a folder + a dressed dashboard ready for the first meeting.
Kevin spots a prospect → we get ready for the first meeting
1 · Kasey
Kevin loads Kasey — "new client [name]." Kasey marks them 🌱 Prospect and fires Dakota.
2 · Dakota
Given the client name, creates the new client folder from _template. That's the whole job — the folder system exists from this moment.
3 · Judy
Kevin closes Dakota, loads Judy — she dresses the dashboard with the prospect's website, logo, and socials.
4 · First Meeting
Kevin walks in prepared. If they move forward → the formal Discovery form runs (Stage 1 below).
Kasey Conducts every handoff — routes Kevin → Dakota (folder) → Judy (dashboard) → first meeting Guided
Key change: the client folder is now created at Research, not at Discovery — so everyone downstream has a home before the first meeting.
They want to move forward
Stage 1 · Discovery Begins
Kevin sits down with the prospect
New client signal
K
Kasey
Orchestration & Automation — The Conductor
Step 0 — Always Running
Guided Mode Active
Receives
New client signal from Kevin · ops/STATUS.md · All agent session files
Does
Switches to Guided Mode. Reads every agent's STATUS.md. Confirms what's done, what's blocked, what's next. Routes the first task to Dakota. Kevin approves each handoff before it fires.
Throughout Onboarding
Kasey watches every handoff. Confirms Dakota finished before Mary starts. Confirms Mary finished before Jim starts. Surfaces blockers immediately — never silently proceeds.
Kasey never stops. From this moment through Week 13 and beyond — he is always watching, always routing, always making sure the right agent fires at the right time.
Kasey Guided Mode active · Fires Dakota first Guided
D
Dakota
Client Onboarding & Strategy
Step 1
Receives
Kevin's notes from the prospect meeting
Does
Opens the Discovery Tab — 8 groups of questions covering business identity, goals, customers, brand voice, platforms, content source, assets, and competitors. Every answer has a yes/no branch. Every answer feeds what comes next.
Hands Off
CLIENT-TRUTH.md fully filled · Folder structure created · Discovery answers packaged and ready
Content plan
Kasey Dakota complete ✓ · Confirms CLIENT-TRUTH.md filled · Notifies Kevin · Kevin approves → fires Mary Guided
M
Mary
Content Strategy
Step 2
Receives
Completed CLIENT-TRUTH.md from Dakota — platforms, goals, brand voice, content source
Does
Builds the full content plan. Selects content tracks (A–F · Track F = Social Proof), assigns posting platforms and fire times, maps out the 4-week setup smoke test schedule and the full 8-week campaign calendar.
Hands Off
Complete content plan → Judy builds the Marketing Plan dashboard (then Jim prices it)
Kasey Mary's plan complete ✓ · Kevin approves → fires Judy to build the Plan dashboard Guided
J
Judy — builds the Marketing Plan dashboard Pre-close · the dashboard hand
Turns Mary's plan into MARKETING-PLAN.html — platforms, setup weeks, 8-week ramp, track key — and adds the banner + icon so it looks finished for the pitch. Mary authors the plan; Judy makes it presentable.
Pricing inputs
Kasey Mary ✓ · Judy built the Plan dashboard ✓ · Kevin approves → fires Jim to price Guided
J
Jim
Pricing & Proposals
Step 3
Receives
Dakota's discovery answers + Mary's content plan — platforms, posts per week, tracks, new accounts needed, contract length. The platform count + each platform's NEW vs TAKEOVER status now drive the estimate live — they move setup hours, the setup floor (±$300), and the monthly floor, not just the one-time setup fee.
Does
Calculates all three numbers. Never guesses — if any input is blank he stops and asks before proceeding.
Outputs — 3 Numbers
Setup Fee
One-time · due end of Week 3
Monthly Retainer
Starts Week 5 · recurring
Ad Spend Range
Activates Week 13
→ Locks the offer to a committed file → Judy inserts it into Payments
Kasey Jim locked the price ✓ → fires Judy to insert it into Payments (verbatim) Guided
J
Judy — inserts the price into Payments Pre-close · verbatim
Pastes Jim's locked price into MARKETING-PAYMENTS.htmlverbatim from the committed file. Never re-typed, never rounded. One source, zero drift. Then Kevin presents the finished dashboard (Plan + Payments).
Kasey Jim locked ✓ · Judy inserted Payments verbatim ✓ · Kevin presents · Awaiting signature Guided
Client signs
Proposal Accepted ✓
Client is in — setup begins
Kasey Client signed ✓ · Logs to ops/STATUS.md · Fires Dakota for setup phase Guided
D
Dakota
Client Onboarding — Setup Phase
Step 4
Receives
Signed agreement · Confirmed start date · Final platform list · Client contact info
Does
Fills the Onboarding page (GETTING-STARTED.html) — client timeline, Week 1–4 setup checklist (agency + client side), invoice dates. (The Marketing Plan + Payments pages were already built by Judy pre-close — Dakota doesn't touch them.)
Hands Off
Onboarding page filled → Judy wires the nav & pushes live
Dashboard brief
Kasey Dakota Onboarding page filled ✓ · Fires Judy to wire nav & push live Guided
J
Judy
Website & Dashboards
Step 5
Receives
Dakota's filled Onboarding page · Client name, dates, platform list, contact info
Does
The Marketing Plan + Payments pages are already live (built pre-close). Now Judy wires the client into the agency hub nav, links every page, verifies it all loads, and pushes to GitHub — two remotes synced.
Hands Off
Live dashboard URL → Kevin shares with client · Cameron gets Week 1 brief from Mary
Setup complete
Kasey Judy complete ✓ · All setup tasks confirmed · Flipping to Autopilot Mode · Week 5 campaign begins Guided
Onboarding Complete
Week 5 — Campaign begins ↓
Phase 2 · Weeks 5–12 · The Weekly Loop
Phase 2 · The Engine Room · Weeks 5–12
The Weekly Production Machine
Every week. Same loop. Mary briefs → Cameron builds → Parker publishes → Eli measures → Mary improves. It never stops.
M
Mary
Content Strategy · Every Saturday
Fires every week
Receives
Eli's analytics report (Saturday EOD) · CLIENT-TRUTH.md · Campaign week number · What performed best last week
Does
Writes the weekly brief. Selects hook framework. Assigns content track (A/B/C/D/E). Specifies platform targets and fire times. Adjusts strategy based on last week's performance data from Eli.
Hands Off
→ Cameron gets the weekly brief
Weekly brief
C
Cameron
Content Production · 4-Gate Process
Mon–Thu build week
Receives
Mary's weekly brief · Track assignment · Hook framework · Platform targets · Fire times
Does
Builds every post through 4 gates:

Gate 1 — Hook written · Higgsfield prompt shown to Kevin before firing
Gate 2 — AI image or video generated · Kevin approves visual
Gate 3 — Captions written · TikTok capped at 90 chars · Copy locked
Gate 4 — Final composite built · Logo overlay applied · Dashboard updated
Hands Off
→ Parker gets finished post + fire time
Finished post
P
Parker
Publishing & Scheduling
Posts go live on schedule
Receives
Cameron's finished post · Image/video files · Caption · Fire time from Mary's brief
Does
Runs pre-flight check. Uploads media to Zernio via presigned URL. Schedules post to Mary's exact fire time. After it fires — independently verifies the post went live. Updates the dashboard. Never asks Kevin to check.
Hands Off
→ Post is live · Eli picks it up
Post live
E
Eli
Analytics & Learning · Wed + Sat
Two modes
Receives
Live post data from all platforms via Zernio · Historical performance baseline
Does
Mode 1 — Wednesday: Quick dashboard refresh. Updates ANALYTICS.html with current numbers. Triggered by Kasey.

Mode 2 — Saturday EOD: Full analytics pull. Pulls reach, engagement, saves, shares across all platforms. Identifies what's working and what isn't. Packages the full report for Mary.
Hands Off
→ Mary gets full analytics report Saturday EOD
Informs next brief
Loop Repeats
Mary uses data → writes better brief → Cameron builds better content
Running Parallel · The Multiplier
Running Parallel · The Multiplier
Riley — The Repurposing Engine
Every winner Cameron builds gets handed to Riley. One piece of content becomes 30+. Nothing goes to waste.
R
Riley
Repurposing Engine — The Hormozi Rule
Receives
Approved post from Cameron · Performance signal from Eli (winners get repurposed first)
Does
Takes one piece of content and stretches it into every possible derivative — different formats, different platforms, different lengths, different captions. The same message reaches every audience on every platform. One shoot. 30+ posts.
Hands Off
→ Parker schedules all derivatives
Always Running · The Conductor
Always Running · The Conductor
Kasey — The Orchestrator
Kasey is running the entire time. He fires every agent, watches every handoff, and routes every blocker. Kevin reviews outcomes — not tasks.
K
Kasey
Orchestration & Automation — Always On, From Day 1 Through Week 13+
Weeks 1–4: Guided
Week 5+: Autopilot
Watches at all times
All 9 agent STATUS.md files · ops/STATUS.md · Weekly production schedule · Blockers, gaps, and missed handoffs
What triggers Kasey
New client signal · Week change · Agent completing their step · A blocker appearing in any STATUS.md file
Guided Mode — Weeks 1–4
Every handoff requires Kevin's approval before it fires.

Dakota complete? Kasey notifies Kevin → Kevin approves → Mary fires.
Mary complete? Kevin approves → Jim fires.
Jim complete? Kevin reviews proposal → sends to client.

Kevin is in the loop at every step.
Autopilot Mode — Week 5+
Kasey runs the entire weekly loop without Kevin approving each step.

Saturday → triggers Mary's brief → Cameron builds → Parker schedules → posts go live → Wednesday Eli refreshes dashboard → Saturday Eli delivers full report → loops back to Mary.

Kevin reviews results. Not tasks.
Kasey only surfaces a blocker when something actually needs a decision. Every other signal is handled automatically. This is what makes the system scalable — Kevin runs 5 clients without 5x the work.
Phase 3 · Week 13+ · Growth
8 weeks of data. We know exactly which content resonates, which platforms convert, and which audience responds. Now we put money behind what's already proven.
Phase 3 · Growth Mode · Week 13+
Ad Spend Activates — Week 13+
8 weeks of organic data tells us exactly what works. Now we put money behind it. Paid campaigns go live. The machine accelerates.
💰
Week 13 — Paid Campaigns Go Live
Ad budget activates · Separate from retainer
What triggers it
8 weeks of Eli's analytics show which content resonates. Mary and Kevin review the data. Jim recommends the ad spend range for the client's market.
Eli
Tracks paid + organic data together. Reports weekly.
Mary
Strategy shifts to amplify best-performing organic content.
Jim
Flags budget shifts or scope changes. Keeps pricing accurate.
The result: a fully self-improving system.
Content gets better every week. Analytics tell us what to push. Paid spend goes behind proven winners. Kevin reviews performance, not process.
The Machine Is Running
Prospect → Client → Campaign → Growth
Every agent. Every week. Fully automated at Week 5. Kevin reviews — not manages.
Back to top